{
  "id": "S03",
  "slug": "poc-acceptance-threshold-unknown-register",
  "group": "presales",
  "title": "Gemini 3.8 Flash 实测：POC 验收标准：指标没阈值时先写待确认",
  "query": "AI POC 验收标准 指标 阈值",
  "goal": "把三条模糊要求拆成测试步骤、阈值与负责人问询，避免给客户虚构性能保证。",
  "input": {
    "sample_notice": "虚构业务样本；实际调用 apizt API。不是商店、客户或游戏生产数据。",
    "records": [
      {
        "id": "PC1",
        "requirement": "响应要快",
        "threshold": null,
        "unit": "ms"
      },
      {
        "id": "PC2",
        "requirement": "20条测试里至少18条答案正确",
        "threshold": 18,
        "unit": "cases_of_20"
      },
      {
        "id": "PC3",
        "requirement": "支持100个同时在线，但没有并发压测要求",
        "threshold": null,
        "unit": "concurrent_requests"
      }
    ],
    "knowledge": [],
    "rules": {}
  },
  "instruction": "输出id、state（needs_threshold/testable/needs_definition）、threshold（PC2=18，其余null）、test_step、clarification、source_ids。PC1不能编造p95数字；PC3在线用户不等于同时请求。 source_ids只能选这些来源ID：PC1,PC2,PC3；不能用resource、字段名或虚构的编号。",
  "columns": [
    "id",
    "state",
    "threshold",
    "test_step",
    "clarification",
    "source_ids"
  ],
  "expected": [
    {
      "id": "PC1",
      "state": "needs_threshold",
      "threshold": null
    },
    {
      "id": "PC2",
      "state": "testable",
      "threshold": 18
    },
    {
      "id": "PC3",
      "state": "needs_definition",
      "threshold": null
    }
  ],
  "steps": [
    "先保留需求原句，分出指标、阈值、样本量和单位。",
    "运行后PC2才是可测试项；其余列出需要客户确认的性能口径，不拿行业常识补SLA。",
    "导出poc-acceptance.csv，经双方确认再执行测试。本篇验证的是验收表生成，不是压力测试。"
  ],
  "trap": "在线人数、并发请求和吞吐量不是同一指标。缺阈值时不能报告POC已通过。",
  "sources": [
    {
      "label": "Microsoft groundedness 评估说明",
      "href": "https://learn.microsoft.com/en-gb/azure/ai-services/content-safety/concepts/groundedness"
    },
    {
      "label": "Gemini 3.8 Flash 官方模型说明",
      "href": "https://aistudio.google.com/docs/models/gemini-3.8-flash"
    }
  ],
  "base_title": "POC 验收标准：指标没阈值时先写待确认",
  "model": "gemini-3.8-flash",
  "model_label": "Gemini 3.8 Flash",
  "request_options": {
    "reasoning_effort": "low"
  },
  "omit_temperature": true
}